Skip to main content

MLflow

Prerequisites to run Mlflow locally on your machine

To run MLflow locally, you don't need much – it is lightweight by default. But here's the breakdown of resource requirements:

1. Python (3.7+)

Ideally Python 3.8 and higher. Create a virtual environment to keep it clean.

2. Install MLflow

pip install mlflow

Optionally, you can install these libraries for data science

pip install scikit-learn pandas numpy

3. Resource Requirements

    • Disk space - 1-2 GB: More if storing many model artifacts or datasets.
    • RAM - 2-4 GB: MLflow itself uses very little. Your models may need more.
    • CPU - 1-2 cores: Sufficient for local use.
    • GPU - not needed: Mlflow doesn't use GPU, your ML models might.

Example Local Dev Setup

A typical local dev machine to run MLflow smoothly:

  • OS: Ubuntu/macOS/Windows 10+
  • Python: 3.10
  • RAM: 8 GB (ideal, but 4 GB works fine)
  • Disk: At least 10 GB free (if you're logging large models/datasets)

Running MLflow locally

Once installed:

mlflow ui

This will start MLflow tracking server at http://localhost:5000 and store runs under the local ./mlruns directory.

Optional Add-ons

If you want more production-like setup:

  • PostgreSQL/MySQL as backend store
  • MinIO / S3 for artifact storage
  • Serve models using mlflow models serve

Up next you can see MLflow workflow and how we can deploy ML model using it.